Russ,

MELODIC by Christian Beckmann and part of FSL has wonderful capabilities to
detect artifacts. See also Christian's little fMRI shop of horror
http://www.fmrib.ox.ac.uk/~beckmann/homepage/academic/littleshop/ . I have
found MELODIC extremely power- & helpful.
Best regards-
Andreas

----- Original Message -----
From: "Russ Poldrack" <[log in to unmask]>
To: <[log in to unmask]>
Sent: Wednesday, April 09, 2003 4:21 PM
Subject: Re: rfx analysis problems


Kathryn,
I would suggest re-running the fixed-effects analysis on those subjects
with the problems, including the motion estimates as covariates in the
analysis.  Also, have you looked at the raw timeseries data to
determine whether there are spikes or other kinds of machine artifacts
in the data?  You might try the AnalyzeMovie tool to view a movie of
your images across each run.

On this note - has anyone written any MR artifact detection routines
that are SPM-friendly?  We have a local tool at UCLA but it does not
work with analyze format images.

cheers
russ

On Wednesday, April 9, 2003, at 01:30 AM, Kathryn Moores wrote:

> Dear SPM list members,
>
> I have performed an rfx analysis with two groups (controls and ptsd)
> with 12 and 13 subjects respectively. This included both within group
> (one-sample t-test) and between group (two-sample t-test) analyses
> using con* images generated from single subject fmri at the first
> level.
>
> To my despair, this analysis has failed on two accounts: (a) the
> within group results are not significant, although they look like
> they might be sensible at a very low uncorrected threshold and (b) the
> between group results are not significant and do not even look like
> they might be sensible (that is, stuff you can see at a low
> uncorrected threshold is not what you would expect based on the within
> group patterns).
>
> In attempting to discover any possible explanation/solution to these
> problems, I have noticed a couple of curious/suspicious things about
> my data (particularly in comparison to the spm sample data set
> -erfmri).
> 1. the maximum values in my single subject ResMS images can be very
> large, in some cases 270,000 compared to about 5,000 (spm data)
> 2. values in the con*.img range from about -30 to +30 whereas the
> con*.img from the example data rangesl from about -1.2 to +1.2.
> These values seem incredibly high and are maximal in the eyes, hence
> my interest in performing explicit masking at the first level. In
> subjects where signal from the eyes was masked out of the analysis
> have much more reasonable maximum variance (aprox 10,000). Has anyone
> had a similar experience with such large residual variance?
>
> Further, about 50% of subjects do not show any significant activation
> for one of my prinipal comparisons (working memory updating vs working
> memory maintenance only). To investigate this I looked at one sided
> con* images (i.e., reflecting the betas for all the blocks associated
> with a specific condition) in these subjects, it appears as though the
> whole brain is activatied in both conditions, which is clearly why
> there is no significant difference between conditions in these
> subjects. The other 50% of subjects show quite normal WM pattern in
> these one sided con*images, and these are the ones that show a
> significant difference. Has anyone else ever experienced this problem?
> I just cannot understand why it would seem the whole brain is
> activated in these subjects?
>
> I am really at a loss to explain or solve the problems I am
> encountering with these analyses. I thought I would probably be okay
> with 12/13 subjects for rfx analysis (although I know more would be
> better, it is very hard to find suitable patients). I also collected a
> substantial amoung of single subject data (1200 volumes per subject)
> to ensure that the results at the single subject level had enough
> power.
>
> If anyone could give me any ideas or leads to investigate, I would be
> most grateful.
> Cheers, Kath
>
> -----------------------------------------------------------------------
> -------------
> Kathryn Moores, PhD Student
>
> Cognitive Neuroscience Laboratory,
> School of Psychology
>
> Flinders University
> Email: [log in to unmask]
> Web:http://www.ssn.flinders.edu.au/psyc/students/KathrynMoores/
> -----------------------------------------------------------------------
> -------------
>
>
>
----
Russell A. Poldrack, Ph.D.
UCLA Department of Psychology
Franz Hall, Box 951563
Los Angeles, CA 90095-1563

email: [log in to unmask]
phone: 310-794-1224
fax: 310-206-5895
web: http://www.poldracklab.org/